Use manifest context limits and handle memory exhaustion safely

Set 8K context defaults and add artifact ZIP export

Use 4096-token release defaults

Fix streamed tool calls and align generation limits

Release v1.2.0 live model telemetry

Release v1.1.3 scored WebGPU evidence

Release v1.1.2 benchmark runtime evidence

Add Phase 5 benchmark and runtime provenance

Release browser-local Bonsai WebGPU chat